That's in the class of things we can't explain without ruining them, but I can at least offer a couple things. First, we're careful not to say anything beyond what we see in the data we have. For sure a sufficiently smart campaign is going to exceed our ability to detect it. But what can one meaningfully say about that? We have to stick to what our flashlight can show, and trust that it is at least more reliable than no flashlight. I can tell you that when it does light on an accusation of astroturfing, it nearly always tends against the accusation.

Second, some of the analysis can be done by anyone who wants to. When you encounter specific claims of astroturfing or shillage, look at the history of the commenter being accused. Most of the time their track record makes it implausible. If someone has been posting to HN for five years including about, say, garbage collection in Julia, what are the odds that they're secretly a foreign agent? Far lower than that the other user tossed off an accusation without pausing to look. It's usually not even a serious question.

Now consider that these demonstrably low-probability cases are exactly like the rest of the accusations people post here, and one has evidence for a common mechanism underlying the entire class. I don't assert (how could I) that there are no cases of genuine manipulation that fall outside it. But after looking at thousands of such claims, I believe that this and similar tests account for nearly all of them.

(Edit: it's different in cases of startups or projects trying to game HN to promote themselves; that's super common. And more sophisticated corporate astroturfing is something we've occasionally run into. But on these political, national, and ideological issues: zilch.)

Interesting, thanks for the response and the work you do.

"Oh no, now all the coordinated disinformation campaigns will start posting five years of high-quality good-faith technical discussion on a variety of topics to evade your heuristic" /s